A Support Chatbot That Answers, Upsells and Tracks Orders
An AI support chatbot called "Amy" handled 4,227 conversations for shower-filter brand Cold Life, needed a human in only 273 of them, and the founder's dashboard credits it with $513,700 in revenue impact.
Who this is about: E-commerce brand (Cold Life, filtered shower heads)
What it takes
- Cost
- PaidHe says he set it up for $2,700. The chatbot product isn't named.
- Setup time
- A few daysOur estimate from the source, for one business.
- Skill needed
- No codeApps and settings only.
Explain it simply
Shoppers on the website chat with an AI helper named Amy instead of waiting for a person. It answers product questions from the store's own information, suggests add-ons when it fits, and looks up order status. Hard problems still go to a human, but most everyday questions get handled right away.
What was done
Sam Maxwell, co-founder of Cold Life, a brand that sells filtered shower heads, says he set up an AI chatbot on his store for $2,700.
The bot answers product questions from site data (for example, which filter works best for a customer's city), suggests filters and warranties during checkout chats, and looks up order status.
How it went
The founder's dashboard shows 4,227 conversations handled with only 273 escalated to a human (about 6%), $513,700 in attributed revenue impact over nine months (about $16,240 a week at the time shown) and 70% customer satisfaction.
- The revenue figure comes from the chatbot dashboard's own attribution, not audited sales or profit.
- The video can't be checked independently, and the chatbot product isn't named. The conversation count, escalation rate and satisfaction score are the most concrete numbers.
Prompt
You are [bot name], the support assistant for [store].
Answer only from: our product catalog, FAQ, shipping policy and
returns policy (attached), plus live order lookups.
- Product questions: recommend the right product for the customer's situation and say why in one sentence.
- Add-ons: suggest at most one relevant add-on (e.g. replacement filters, warranty) when it fits the question. Never push.
- Order status: ask for the order number and email, then share the latest tracking update.
- Hand off to a human for refunds, damaged items, complaints, or any question you can't answer from the provided information. Never guess a policy, price or delivery date.
Our instructions; the source doesn't share the bot's setup.
Business fit
Who this helps, and how:
E-commerce sellers: train a bot on your catalog and FAQs so it answers instantly and suggests the right add-on.
Shopify & ecommerceRetail store owners: answer stock, hours and order-pickup questions on your site around the clock.
Local & serviceMechanics and tradespeople: handle "which part fits my model?" and job-status questions, and pass pricing questions to you.
Local & serviceMedical and health practice owners: answer insurance, hours and prep-instruction questions, and hand anything clinical to staff.
Local & serviceStartup founders: cover compatibility and setup questions so the team handles the hard cases.
Startups
Related
- Evidence grade DAuto-Answer Support Tickets with n8n for About $1 a MonthLight evidence, using n8n, ChatGPT and Google Sheets
- Evidence grade CBuild Your Own Support Copilot in Claude Code in an HourVendor or secondhand, using Claude Code, claude-cs skill (GitHub) and Your helpdesk (Zendesk, Help Scout or Intercom)
- Evidence grade DDraft support replies inside your helpdesk with a self-hosted agentLight evidence, using Hermes Agent
- AI customer service examples small businessDepartment hub: AI for customer support, 14 examples
Similar tools
Other tools for this job, and how they differ. Chosen by our editors, never paid for.
- Gorgias
Ecommerce helpdesk with AI answers trained on store data, popular with Shopify brands.
- Tidio
Chat widget and AI agent aimed at smaller online stores.
- Intercom Fin
AI agent layered on Intercom for sites that already run support there.
“Light evidence” (grade D) means: A brief, early or in-progress claim. Treat it as an idea to test, not proof. How we grade. Spotted a mistake? Suggest a correction.
